Comparative Performance of Single Layer and Multilayer Microwave Filtersincluding the Influence of the Fabrication Process

Wesam Ali, Chunwei Min, and Charles FreeAdvanced Technology Institute, University of Surrey

Ouildford Surrey OU2 7XH, UKW.Ali@surrey.ac.uk

517

AbstractModem fabrication processes offer the microwave

circuit designer a wide range of techniques and materialsfor realizing passive planar components, such as band­pass filters. This paper investigates the effects that thefabrication process has on the performance of microwaveplanar circuits, so as to provide the circuit designer withuseful practical information leading to the optimumchoice ofmaterials and processes. In order to compare thetechnologies, single-layer and multi-layer band-pass filtercircuits were fabricated and tested at 250Hz and 2.50Hz,respectively.

IntroductionEdge-coupled planar band-pass filters were chosen for

this study because they are particularly sensitive to errorsin the fabrication process [1]. Notably, in single-layerdesigns, errors in the gap size between the end sections ofthe filter have a significant effect on the electricalperformance of the filter. The problem associated withfabricating small coupling gaps in a single layer circuitcan be overcome by using a multi-layer structure, wherethe coupling is between overlapping conductors separatedby dielectric. In a filter, this produces strong couplingbetween the elements of the filter, without the need forsmall gaps [2]. To some extent, the problem of fabricatingsmall gaps has been exchanged for that of achieving highlateral resolution between the layers. However, withmodem mask aligners it is relatively straightforward toachieve the required degree of resolution.

The two circuit technologies, namely polymer andphotoimageable thick film chosen for the present studyenable both single and multi-layer circuits to befabricated. The photoimageable thick-film processenables the layers to be built up successively using aprint-and-fITe process. The polymer approach enables thecircuit layers to be processed in parallel and finallystacked and laminated under relatively low temperatureand pressure. In many ways this polymer technology is acompetitor to the rather better established LTCC (lowtemperature co-fired ceramic) process, and offerspotential advantages in terms of low-cost production.

Included in the paper is a design strategy formultilayer, edge-coupled, bandpass filters (ECBPFs). Insingle-layer structures, such as microstrip and stripline,ECBPFs are relatively easy to implement with workingbandwidths up to 15-20%. Wider bandwidths requiretighter coupling between conductors, and this is wheremultilayer structures offer significant benefits.

978-1-4244-2814-4/08/$25.00 ©20081EEE

Measurement data for an ECBPF multilayer design arepresented, to validate the proposed design methodology.

Design of ECBPF on Polymer SubstratePolymer substrates are normally supplied from the

manufacturer pre-coated with thin copper conductor,which can be etched to form the required patterns using aconventional, high-quality PCB process.

In this study, a 4-section bandpass filter was designedwith a centre frequency of 250Hz with a fractionalbandwidth of 10%, and a roll-off specified by IS2112<­30dB at 200Hz. The filter was fabricated on polymersubstrate (cr = 2.5) having a thickness of 0.128mm. Thephysical dimensions of each of the coupled sections areshown in Table I.

TABLE IPHYSICAL DIMENSIONS OF THE FILTER ON

POLYMER SUBSTRATE AT 250Hz

Section 1 2 3 4

Width (W) 0.205 0.314 0.314 0.205

Space (S) 0.018 0.071 0.071 0.018

Length (L) 2.083 2.031 2.031 2.083(in mm)

After fabrication the filter was characterized using anHP 8510 vector network analyzer, with the filter mountedin a Wiltron Universal Test Fixture. Fig. 1 shows thesimulated and measured responses of the designed filter.The simulated filter had a centre frequency of25.065GHzwith a 3dB bandwidth of3.970Hz (15.8%), and a roll-offof IS2112<-30dB at 200Hz. The maximum insertion loss inthe passband is around 1.77dB. The measured circuit hada centre frequency of 25.45 OHz with a 3dB bandwidthof 3.650Hz (14.34%), and a roll off of IS2112<-30dB at21.80Hz. Thus it is seen that the centre frequency of thefabricated filter has shifted by 1.53%, with the bandwidth1.46% less than the desired value.

To provide some traceability of the errors due to thefabrication process, mask and circuit dimensions weremeasured at various stages in the process. Table II showsthe physical dimensions of the design, the dimensions onthe mask and the dimensions of the filter after thefabrication. It is clear from Table II that the physicaldimensions of the design almost match with thedimensions on the mask while there is a differencebetween the designed values and the fabricated circuitdimensions. The width and the length of each filter

2nd Electronics Systemintegration Technology ConferenceGreenwich, UK

Authorized licensed use limited to: University of Surrey. Downloaded on March 17,2010 at 07:29:45 EDT from IEEE Xplore. Restrictions apply.

(in mm)

section decreased after the etching process while thespacings between each filter section became larger.

TABLE IIPHYSICAL DIMENSIONS OF THE FILTER ON

POLYMER AT 25GHz

mm)(in

TABLE IIIPHYSICAL DIMENSIONS OF THE FILTER ON

ALUMINA SUBSTRATE AT 25GHz

effect of undercutting of the metallization was probablythe main reason for the reduction of the measuredbandwidth, and for the poor return loss. The latterquantity is very dependent on the conductor widths andspacings being correct so as to maintain the correct oddand even mode impedances. The losses due to thedielectric material and bulk conductors had been includedin the simulation, although there may be additionalinsertion loss due to the surface roughness of the fmalcircuits.

Design of ECBPF on Alumina SubstrateFor this design, an edge-coupled band-pass filter in

single layer format was fabricated on typical 96%alumina substrate having a thickness of 0.254mm and arelative permittivity (E:r) of 9.8. A photoimageable thick­film (PTF) process was used to realize the conductorpattern. The filter had been designed with a centrefrequency of 25GHz with a required fractional bandwidthof 10%, and a roll-off specified by IS211<-30dB at20GHz. The physical dimensions of each of the coupledsections are shown in Table III.

Section 1 2 3 4

Width (W) 0.127 0.210 0.210 0.127Space (S) 0.066 0.203 0.203 0.066

Length (L) 1.130 1.078 1.078 1.130

- MealS111- MealS21I-+- SimlS111-+- SimlS21I

-20

-25

CD -15"C

-30 ...........__---'-__....s.-__-'---_~

20 22 24 26 28 30Frequency (GHz)

Fig. 1. Simulated and measured responses of the filter design onpolymer at 25GHz.

Section 1 2 3 4W - Design 0.205 0.314 0.314 0.205

W - Mask 0.205 0.214 0.214 0.205

W - Fabricated 0.160 0.262 0.262 0.160

S - Design 0.018 0.071 0.071 0.018S - Mask 0.017 0.0.07 0.070 0.017

S - Fabricated 0.054 0.121 0.121 0.054

L - Design 2.083 2.031 2.031 2.083

L - Mask 2.080 2.006 2.006 2.080L - Fabricated 2.072 1.928 1.928 2.072

-40 '----~------------------'20 22 24 26 28 30

Frequency (GHz)Fig. 3. Simulation of filter insertion loss showing effects ofvarious percentage errors in the conductor widths and spacings.

Fig. 2 illustrates the effect of undercutting in thefabrication process, and clearly this is going to be asignificant issue in realizing filters with high, predictableperformance.

:+-w--+: :I ..-5--...I I II I I

Fig. 2. The effective dimensions after the etching process.

-10

-30

/,/,I

/I,l

//'/

,,//

////+10%

- Oak_ .... -100k

If we assume an etching ratio of unity, the followingdesign equations in (I) may be used to compensate for thephenomenon of undercutting by increasing the width (W)of the conductors on the photo-mask

W=Wd+2t S=Sd-2t L=Ld +2t (I)

where Wd, Sd, Ld, are the design dimensions, and t is thethickness of the metal (copper) used in the circuit. Themeasured thickness of the copper was 0.017mm. The

Due to the variations in circuit geometries that canoccur during thick film processing, analysis of the effectson the fabrication errors is essential. Simulation datashown in Fig. 3 and Fig. 4 indicate how the insertion lossand matching of the filter vary with ±10% change incircuit dimensions. A 10% decrease of the length andwidth of the filter elements results in a narrowing of thebandwidth of about 15.12% with worse roll-off, and aslight downward shift of the centre frequency of about

5182nd Electronics Systemintegration Technology Conference

Greenwich, UK

Authorized licensed use limited to: University of Surrey. Downloaded on March 17,2010 at 07:29:45 EDT from IEEE Xplore. Restrictions apply.

Ii1:

tj\:Iti

Multilayer ECBPF DesignFig. 6 shows configuration of the proposed multilayer

microstrip coupled-line filter. The coupling was achievedbetween asymmetrical conductors on different layers,with the configuration shown in Fig. 7.

o--;::--:=---~_ ...-----t-10

-20

en -30"C

-40

-50

-6°20 3525 30Frequency (GHz)

Fig. 5. Simulated and measured responses of the filterdesign on alumina at 25GHz.

Embedded Layer Wemb 1... _Fig. 6. Configuration of the proposed multilayer microstripcoupled-line filter.

the dielectric. This effect becomes increasinglysignificant at mm-wave frequencies.

Fig. 7. Asymmetric microstrip coupled-line structure.

When the coupled lines are positioned on differentdielectric layers they become physically asymmetricbecause of the need to maintain identical line impedances,with the same propagation characteristics. Hence theselines no longer support pure even and odd modes. Rather,the two excitation modes become dependent,corresponding to the c- and n-modes. A full-wave basedanalysis in conjunction with the specified modal

+10%-- 0%

-100/0

0..-......_~=::::".""--r----r----...~_1!!JIIIlI---,

-10

-20

al-30"C

-40

-50-60 L-_---&..__--'--__....a.J.-. ...I.---J

20 22 24 26 28 30Frequency (GHz)

Fig. 4. Simulated return loss results for different fabricationerrors.

4.41 %. A 10% of increase of dimensions results in awider bandwidth of about 18.28% with better roll-off,and slightly upward shift of the centre frequency ofaround 4%. This analysis gives a useful insight into theeffect of shrinkage during the firing process on the filter'selectrical performance, and clearly shows the need for theeffects of the fabrication process to be considered at thedesign stage.

Fig. 5 shows a comparison between the simulated andmeasured responses of the filter. The filter had a centrefrequency of25.135 GHz with the 3dB bandwidth about3.99GHz (15.87%) and the roll-off of IIS2112<-27dB at20GHz. The maximum insertion loss in the passband isaround 0.031 dB. The return loss is below 15dB whichrepresents good matching of the circuit around theoperating frequency band. The measured circuit had acentre frequency of 26.75GHz with 3dB bandwidthwhich to be about 3.5GHz (13.08%) and roll off ofIS21121<-30dB at 22GHz with insertion loss of 6.06dB. Thereturn loss is below 10dB which represents reasonablygood matching of the circuit around the operatingfrequency band and meets the required specification. It isseen from the results that the centre frequency shifts by1.11% and the bandwidth reduces slightly by 2.79%.

It is seen from the results that the bandwidth reducesslightly and this was attributed to shrinkage of theconductors during fIring, which resulted in an increase inthe spacing of the conductors in each section of the filter.The slight shift in the centre frequency may have beendue to shrinkage effects, but more likely to incorrectcompensation for electromagnetic fringing at the end ofeach section. Although the losses due to the dielectricmaterial and bulk conductors had been included in thesimulation, the additional insertion loss may be due to thesurface roughness of the frred circuits. It was observedthat the insertion loss was higher for the thick-film filters,compared to those made on polymer, where the copperconductor was laminated onto the polymer. In thesecircumstances the thick-film surface would be expected tobe somewhat rougher than that of the copper. This agreeswith results published elsewhere [3] that show that theconductor loss at these frequencies is greater than that in

2nd Electronics Systemintegration Technology Conference519 Greenwich, UK

Authorized licensed use limited to: University of Surrey. Downloaded on March 17,2010 at 07:29:45 EDT from IEEE Xplore. Restrictions apply.

parameters enabled a procedure for the filter design to bedeveloped. There are three main electrical parametersused for the design of coupled-line structures, namely,modal impedance, effective dielectric constant, andmutual coupling coefficient. Modal impedances (ZC,1r)decide the corresponding widths for the top andembedded lines, which can be obtained using equation(2).

result. The results also emphasized the wide bandwidthspossible with the multilayer configuration. The mismatchwithin the passband may be due to errors in thefabrication process, including layer-to-layer registrationerrors. These effects become pronounced when theoperating frequency of the filter is higher, extending intothe mm-wave band.

(2)

The electrical length (0) of the coupled lines istheoretically a quarter guide-wavelength (Agl4, n12) long.Open-end effects of this structure can be represented byan additional length (/:4) of the line that needs to besubtracted from the nominal Ag/2 length. The physicallengths (L) of the lines used in the filter design are thusobtained by calculating the effective dielectric constant(ce.t.lJ)) at the frequency of interest through the followingequation

-10

!g -20

-30

2 2.5 3FrequencYt GHz

Fig. 8. Responses of the filter design at 2.5GHz.

3.5

where c is the speed of light.The coupling factor (kc) may be obtained according to

the modal impedances appropriately chosen for the twolines using

(z -Z Jk - 20log ct,e m,ec(dB) -

Zct e + Zm e, ,

TABLE IVDIMENSIONS OF THE DESIGN AT 2.5GHz

(3)

(4)

ConclusionsThe work has identified those aspects of the

fabrication process that have most influence on theperformance of a microwave circuit. Furthermore, thework has indicated that selecting substrate and conductormaterials consideration must be given to the limitations ofthe associated fabrication process.

A design procedure for edge-coupled multi-layerfilters has been developed, and validated throughpractical measurement. The results for the multi-layerfilter have reinforced the benefits of this type of circuitstructure, particularly in overcoming the limitations ofconventional, single-layer components. Using the multi­layer formats, high quality components with widebandwidths (> 20% in our case) are possible with low­cost fabrication processes.

Sec., n 1 2 3 4

WtOlJ 2.610 3.110 3.110 2.610

LtolJ 20.950 20.850 20.850 20.950

Wemb 1.180 1.430 1.430 1.180

Lemb 19.800 19.830 19.830 19.800

S -0.700 0.140 0.140 -0.700

(in mm)

The proposed filter designs were realized usingRT/duroid® 5880 material with Cr}=cr2=2.2 andh}=h2=0.508mm. The design frequency was 2.5GHz.Table IV summarizes the physical dimensions of thefilters at 2.5GHz and Fig. 8 shows the simulated andmeasured responses. The measured bandwidth andinsertion loss of the design were found to be 20.4% and0.6dB, which was in close agreement with the simulated

520

References1. Edwards, T. C., and Steer, M. B. , Foundations of

Interconnect and Microstrip Design, J. Wiley & Sons,3rd ed., (2000).

2. Tsai, C.-M., and Gupta, K. C., "A Generelized Modelfor Coupled Lines and Its Applications to Two-LayerPlanar Circuits," IEEE Tranactions on MicrowaveTheory & Technique, Vol. 40, No. 12 (1992), pp.2190-2199.

3. Henry, M., and Free, C., "Electrical Charecterizationof LTCC Coplanar Lines up to 110GHz," EuropeanMicrowave Con/, Manchester, UK, 2006.

4. Pozar, D. M., Microwave Engineering, J. Wiley &Sons, 3rd ed., (2005).

5. Gupta, K. C., Garg, R., Bahl, I., and Bhartia, P.,Microstrip Lines and Slotlines, Artech House, (1996).

2nd Electronics Systemintegration Technology ConferenceGreenwich, UK

Authorized licensed use limited to: University of Surrey. Downloaded on March 17,2010 at 07:29:45 EDT from IEEE Xplore. Restrictions apply.